The dynamics of body centre of mass (BCoM) 3D trajectory during locomotion is crucial to the mechanical understanding of the different gaits. Forward Dynamics (FD) obtains BCoM motion from ground reaction forces while Inverse Dynamics (ID) estimates BCoM position and speed from motion capture of body segments. These two techniques are widely used by the literature on the estimation of BCoM. Despite the specific pros and cons of both methods, FD is less biased and considered as the golden standard, while ID estimates strongly depend on the segmental model adopted to schematically represent the moving body. In these experiments a single subject walked, ran, (uni- and bi-laterally) skipped and race-walked at a wide range of speeds on a treadmill with force sensors underneath. In all conditions a simultaneous motion capture (8 cameras, 36 markers) took place. 3D BCoM trajectories computed according to 5 marker set models of ID have been compared to the one obtained by FD on the same (about 2700) strides. Such a comparison aims to check the validity of the investigated models to capture the 'true' dynamics of gaits in terms of distance between paths, mechanical external work and energy recovery. Results allow to conclude that: 1) among gaits, race walking is the most critical in being described by ID, 2) among the investigated segmental models, those capturing the motion of 4 limbs and trunk more closely reproduce the subtle temporal and spatial changes of BCoM trajectory within the strides of most gaits, 3) FD-ID discrepancy in external work is speed dependent within a gait in the most unsuccessful models, and 4) the internal work is not affected by the difference in BCoM estimates.